Skip to content

Can Mac M3 Ultra 192G run Kolibri-1 78B-A3.5B?

Comfortable: GGUF Q8_0 needs 87.1 GB at 4K, batch 1; headroom 56.9 GB.

Text input, FP16 KV cache, one card. Unified-memory capacity uses the GPU guide’s usable limit; negative headroom may require host offload. Speed source settings may differ from this 4K estimate.

Measured runs

No measured run for this pair. Sizing is an estimate, not a runtime test.

Quant memory at 4K

QuantTotal GBVerdictWeight sizing
GGUF IQ1_M20.9ComfortableVerified file
GGUF IQ2_XXS24.0ComfortableVerified file
GGUF IQ2_XS25.6ComfortableVerified file
GGUF IQ3_XXS35.4ComfortableVerified file
GGUF Q3_K_M39.5ComfortableVerified file
GGUF IQ4_XS45.8ComfortableVerified file
GGUF Q4_K_M52.2ComfortableVerified file
GGUF Q8_087.1ComfortableVerified file

Context capacity · GGUF Q8_0

ContextTotal GBVerdict
4K87.1Comfortable
16K87.4Comfortable
32K87.7Comfortable
128K89.8Comfortable

Generation speed

GGUF Q8_0: 231 tok/s · Bandwidth ceiling (not measured)

Other model / GPU pairs